Journal of Theoretical and Applied Information Technology
th
10 September 2014. Vol. 67 No.1
© 2005 - 2014 JATIT & LLS. All rights reserved.
ISSN: 1992-8645
www.jatit.org
E-ISSN: 1817-3195
INTELLIGENT SEGMENTATION OF MR BRAIN
STRUCTURES USING SUPERVISED CONTEXTUAL
CLUSTERING
1
S.P. RENJITH, 2DR. N. KESAVAN NAIR
1
Research scholar, Anna University, India
2
Professor, CSI Institute of Technology, India
ABSTRACT
Image segmentation plays a vital role in medical imaging applications. Developing a robust and efficient
algorithm for medical image segmentation has been a demanding area of growing research of interest
during the last two decades. Image processing techniques provide a good tool for improving the manual
screening of samples of Brain. This study presents a novel intelligent approach for the segmentation of
anatomical brain structures in Magnetic Resonance Images (MRI) using supervised contextual clustering
method. Matlab software ‘region props’ function has been used as one of the criteria to show the
performance of the proposed segmentation method. The CC segmentation shows more segmented regions
with less discontinuity within the objects present in the Brain image. The segmented results are compared
with the conventional algorithms
Keywords: Image Segmentation, Medical Imaging, Brain Image, CC Segmentation, MRI
1. INTRODUCTION
A novel method for segmenting the region of
interest present in the MRI brain images was
proposed by Jun Kong et al in [8]. In their
method, initially they reduced the noise present
in it using a versatile wavelet-based filter. Then a
watershed algorithm is applied to the brain
tissues as an initial segmenting method in which
the white matter along with some outer tissues
are segmented. Inorder to reduce the outer
tissues, further, they have used fuzzy clustering
algorithm which removes the over segmentation.
Segmentation of tissues present in the medical
images is an initial step in many medical
imaging applications which includes three
dimensional volume visualization, the detection
of pathology and structure-functions present in it
inorder to map both the scientific and clinical
investigation.
Accurate
segmentation
of
magnetic resonance images (MRI) is essential in
many
neuroimaging
applications
[1].
Quantitative analysis of anatomical brain tissue
such as white matter (WM), gray matter (GM)
and cerebrospinal fluid (CSF) is important for
clinical diagnosis, therapy of neurological
diseases and for visualization and analysis [2].
Because of the advantages of MRI over other
diagnosis of images, many researches in medical
image segmentation applied its use for MR
images. The different types of MR images
obtained from the different excitation sequences,
also called multispectral images, can provide
different image intensity information for a given
anatomical region and subject [7]. As we know,
manual segmentation of MR brain images by an
expert for all types of MR modalities involved in
studies is not only exceedingly time consuming,
but also exhausting for experts, which can lead to
human errors. Therefore, an automatic
segmentation method is necessary. Various
methods are available for MRI image
segmentation which are present in the literature
[2-7].
An improved method for segmenting the ROI of
MRI brain was proposed by Arnaldo et al in [9]
which segments the brain image based on an
adaptive mean-shift methodology in order to
classify brain voxels in to one of three main
tissue types such as gray matter, white matter,
and Cerebro-spinal fluid. The fast adaptive
mean-shift algorithm (FAMS) is utilized to
analyze the MRI brain data inorder to provide
the segmentation maps of the three main tissue
types. Then the categorization of the above
mentioned resultant models is achieved through
an intensity-based clustering stage. Experimental
results revealed that their proposed model gives
more segmentation accuracy in segmenting the
three tissue types. Combining the spatial
information and the statistical atlas paves a way
for improving the segmentation results [10],
[11].
145
Journal of Theoretical and Applied Information Technology
th
10 September 2014. Vol. 67 No.1
© 2005 - 2014 JATIT & LLS. All rights reserved.
ISSN: 1992-8645
www.jatit.org
E-ISSN: 1817-3195
statistical clustering techniques in image
segmentation was inspired by the statistical
based methods. These methods were developed
to study the equilibrium properties of large,
lattice based systems consisting of interacting
components as identical. In a clustering
technique for image segmentation, each pixel is
associated with one of the finite number of
categories to form disjoint regions.
In [11], the intensity of each and every
tissue present in the brain image is modeled by a
Parzen density fitted in the voxels which were
selected from an affine-registered atlas. It is
important stress that each and every atlas does
not exist for the data at hand. Example of two
such cases are brain data with pathologies or
brain data obtained from young infants without
any pathologies. In addition, conventional
method improves the segmentation smoothness
inorder to model the neighboring voxels using a
Markov random field (MRF) based statistical
spatial model [12]. These MRF-based algorithms
are computationally intensive in nature requiring
critical settings in its parameter. It is possible to
use MRF with predefined settings, which is
computationally faster but possibly less in
accurate.
This paper is organized as follows.
Section II explains an overview of the materials
and methods present in this image segmentation
process. Section III presents the proposed
methodology in detail. Section IV provides the
experimental results, and Section V contains the
conclusion on this work and suggests some
possible future enhancements.
The contextual clustering based algorithms are
assumed to be drawn from standard normal
distribution. It segments a data into category 1
(ω0) and category 2 (ω1).
The following are the steps adopted for
implementing the contextual clustering algorithm
for segmenting the white matter from the Brain
MRI images.
(i)
(ii)
(iii)
2. MATERIALS AND METHODS
The primary objective of this work is to
develop a Computer Aided System for
Segmentation of white matter from the Magnetic
Resonance images (MRI). Our proposed method
does not affects the objects present in the lung
image. The main advantage of this approach is
that we can get the holes and objects present in
the original image after segmentation without
any change in its size and shape. This proposed
approach has two main sections namely 1)
Segmentation based on Contextual clustering
and 2) Objects detection and estimation based on
Matlab region properties.
(iv)
(v)
Define decision parameter Tcc (positive)
and weight of neighborhood information β
(positive). Let Nn be the total number of
data in the neighborhood. Let Zi be the
data itself, i.
Classify data with zi>Tα to ω1 and data to
ω0. Store the classification to C0 and C1.
For each data i, count the number of data
ui, belonging to class ω1 in the
neighborhood of data i. Assume that the
data outside the range belong to ω0.
Classify data with
to ω1 and other data to ω0. Store the
classification to variable C2.
If C2 ≠C1 and C2 ≠ C0, copy C1 to C0, C2
to C1 and return to step iii, otherwise stop
and return to C2.
The contextual clustering implementation is as
follows:
Step 1: Read a Pattern (Brain image feature).
Step 2: Sort the values of the Pattern.
Step 3: Find the Median of the Pattern Cm.
Step 4: Find the number of values greater than
the Median Values, Um.
Step 5: Calculate CC using Cm + (beta/Tcc) *
(Um – (bs/2)).
Step 6: Assign CC as the segmented values.
2.1 Contextual Clustering Based
Segmentation
Image segmentation plays an important
role in image analysis and computer vision and it
is considered as one of the major obstruction in
the development of image processing
technology.
Recently,
there
has
been
considerable interest among researchers in
146
Journal of Theoretical and Applied Information Technology
th
10 September 2014. Vol. 67 No.1
© 2005 - 2014 JATIT & LLS. All rights reserved.
ISSN: 1992-8645
www.jatit.org
E-ISSN: 1817-3195
12
Error between estimated
and target value
14
12
Error between estimated
and target value
14
10
8
6
4
2
0
10
8
6
4
2
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10 20 30 40 50 60 70 80 90 100
Threshold
Beta value
Fig. 1 Impact Of Beta Value In CC Estimation
Fig. 2 Impact Of Threshold Value In CC Estimation
Figure 1 shows a steady increase in the error as the beta value changes from 0.1 to 1. Hence, lower beta
value is preferred for better estimation by CC. Figure 2 shows a steady decrease in the error as the
threshold value changes from 10 to 100. Hence, higher threshold value is preferred for better estimation by
CC.
3. EXPERIMENTAL RESULTS AND
DISCUSSION
Brain MRI images have been
considered in this presentation. Magnetic
resonance images of different patients have been
considered. The CC segmented results along
with its original images has been presented in
Table 1
Table 1 Segmented Results Of Different Brain Datasets
No
Brain Image
Segmented result
1
147
Journal of Theoretical and Applied Information Technology
th
10 September 2014. Vol. 67 No.1
© 2005 - 2014 JATIT & LLS. All rights reserved.
ISSN: 1992-8645
www.jatit.org
2
3
4
5
148
E-ISSN: 1817-3195
Journal of Theoretical and Applied Information Technology
th
10 September 2014. Vol. 67 No.1
© 2005 - 2014 JATIT & LLS. All rights reserved.
ISSN: 1992-8645
www.jatit.org
Except CC method, in all other segmentation
methods, the number of objects are more and
there are some objects segmented are not clear.
Matlab ‘bwlabel’ function has been used and the
number objects for each method is shown in
E-ISSN: 1817-3195
Table 1. In addition to bwlabel’, the ‘Region
props’ command has been used to find out
correct number of segmented objects.
6000
No of objects
5000
4000
3000
2000
1000
0
Original
Sobel
Prewitt
Roberts
Log
Zerocross
CC
Segmentation methods
Figure 3: Comparison Of Number Of Objects In Each Segmentation Methods
Earlier researchers had used different metrics to
evaluate the segmentation accuracy. In this
paper, we have used ‘bwlabel’ and ‘Region
props’ of Matlab to evaluate the accuracy of
segmentation and from the experimental results,
it has been found that Contextual Clustering
based segmentation gives much better
segmentation accuracy when compared to that of
remaining segmentation methods mentioned in
this work.
Total Areaof all objects , pixels
30000
25000
20000
15000
10000
5000
0
Original
Sobel
Prewitt
Roberts
Log
Zerocross
CC
Segmentation methods
Figure 4. Comparison Of Total Area Of Pixels In Each Segmentation Methods
Table 1 Segmentation Accuracy For Different Methods
No
1
2
3
4
5
6
Method
Sobel
Prewitt
Robertz
Log
Zerocross
Contextual Clustering
149
Objects detected
1270
84
47
80
74
14
Journal of Theoretical and Applied Information Technology
th
10 September 2014. Vol. 67 No.1
© 2005 - 2014 JATIT & LLS. All rights reserved.
ISSN: 1992-8645
www.jatit.org
E-ISSN: 1817-3195
MR images”, Image and Vision
Computing, vol 25, pp- 164–171, 2007.
[8] Jun Kong, Jianzhong Wang, Yinghua Lu, “A
Novel Approach for Segmentation of MRI
Brain Images”, IEEE MELECON 2006,
pp- 525-528.
[9] Arnaldo Mayer* and Hayit Greenspan, “An
Adaptive Mean-Shift Framework for MRI
Brain
Segmentation”,
IEEE
TRANSACTIONS
ON
MEDICAL
IMAGING, VOL. 28, NO. 8, 2009, pp1238-1250.
[10] P. Anbeek, K. L. Vincken, G. S. van
Bochove, M. J. P. van Osch, and J. van der
Grond, “Probabilistic segmentation of brain
tissue in MR imaging,” NeuroImage, vol.
27, pp. 795–804, 2005.
[11] S. K. Warfield, M. Kaus, F. A. Jolesz, and
R. Kikinis, “Adaptive, template moderated,
spatially varying statistical classification,”
Med. Image Anal., vol. 4, pp. 43–55, 2000.
[12] Y. Zhang, M. Brady, and S. Smith,
“Segmentation of brain MR images
through a hidden markov random field
model and the expectation maximization
algorithm,” IEEE Trans. Med. Imag., vol.
20, no. 1, pp. 45–57, Jan. 2001.
4. CONCLUSION
In this paper we have proposed a new method for
segmenting brain MR images based on
supervised contextual clustering method. The
main purpose of proposing this contextual
clustering based segmentation method is to
improve the segmentation accuracy by reducing
the false segmentation. The method was
validated both on simulated and real brain
datasets, and the results were compared with
state-of-the-art algorithms. We also represented
that using the proposed framework, segmentation
of the normal tissues is not degraded by the
presence of abnormal tissues. Our proposed
approach has been found more efficient when
compared with other conventional segmentation
algorithms. Matlab region properties are used for
the estimation of the white matter segmentation
accuracy.
REFERENCES
[1]. R. Larsen, M. Nielsen, and J. Sporring,
“Atlas Guided Identification of Brain
Structures by Combining 3D Segmentation
and SVM Classification”MICCAI-2006,
pp. 209–216, 2006
[2]. Pham, D., Xu, C., Prince, Current methods
in medical image segmentation. Annual
Review of Biomedical Engineering 2
(2000) 315–337
[3] Alan Wee-Chung Liew, Hong Yan: An
Adaptive Spatial
Fuzzy
Clustering
Algorithm
for
3-D
MR
Image
Segmentation, IEEE Transaction on
Medical Imaging, vol 22, No 9 (2003).
[4] Chris A. Cocosco, Alex P. Zijdenbos, Alan
C. Evans: A fully automatic and robust
brain MRI tissue classification method.
IEEE Transaction on Medical Image
Analysis, vol.7, (2003) 513-527.
[5] K.V. Leemput, F. Maes, D. Vandermeulen,
P. Suetens, A unifying framework for
partial volume segmentation of brain MR
images, IEEE Trans. Med. Imaging 22
(2003) 105–119.
[6] V. Barra, J. Boire, Automatic segmentation
of subcortical brain structures in MR
images using information fusion, IEEE
Trans. Med. Imaging 20 (2001) 549–558.
[7] Weibei Dou, Su Ruan, Yanping Chen ,
Daniel Bloyet , Jean-Marc Constans,
“framework of fuzzy information fusion for
the segmentation of brain tumor tissues on
150